From: Derek G. <fri...@gm...> - 2010-01-14 21:28:45
|
On Jan 14, 2010, at 1:40 PM, Jed Brown wrote: > Yeah, I try to avoid assembly in these cases, by lagging the > preconditioner and doing more things matrix-free (the matrix-free > operations can certainly benefit from threading, you'll probably realize > a greater speedup this way since you can parallelize the mat-vecs, which > are CPU-bound instead of memory-bound when done unassembled -- usually > by storing the nonlinearity at quadrature points). BTW we do everything using JFNK (Jacobian Free Newton Krylov)... and heavily use threads... and see some good gains (as you say you should). And of course we're using libMesh for everything. We're actually in the middle of putting together our talk for next month's SIAM Parallel Processing conference where we're on the hook to give a talk entitled: "Hybrid Parallelism for Preconditioned Jacobian-Free Newton-Krylov Multiphysics Applications". After the conference I can put the slides somewhere for interested parties. Derek |